Comparative Assessment of Deep Learning to Detect the Leaf Diseases of Potato based on Data Augmentation

Abstract

In recent times, the Convolution Neural Networks (CNNs) is widely used in agriculture fields such as plant disease detection, plant health issue prediction, etc. This paper also forwards a self-build CNN (SBCNN) for potato disease detection. The SBCNN is separately applied in the augmented and non-augmented potato leaf image dataset. The algorithm is used to train and test the potato leaves images. The best validation accuracy of SBCNN in the non-augmented and augmented datasets is 96.98% and 96.75% with the training accuracy of 99.71% and 98.75%, respectively. The errors of training and validation are reported in each epoch. The SBCNN model is performed well in an augmented dataset without having any overfitting in the model. The model is also compared with the performance of MobileNet architecture for the development of smartphone applications. Finally, the SBCNN (Augmented) is selected as the best model as compared to SBCNN (non-augmented) and MobileNet. The model is deployed in an android application for real-time testing of potato leaf diseases and it can be considered as a replica of agriculture pathological laboratory.